SR/SSD 98-36


Technical Attachment


Capt. R. Tom Tibbetts, USAF

Cooperative Institute for Tropical Meteorology

Dept. of Meteorology, Florida State University


After three months of extreme heat and unprecedented drought throughout the state of Florida, afternoon thunderstorms returned to north Florida with a vengeance on July 13, 1998. An active line of organized convection from the west collided in the vicinity of Tallahassee, Florida, with a separate line of thunderstorms pushing from the east. In just a few short hours, 4.43 inches of precipitation filled the rain gauge at the Tallahassee airport. Other locations, such as the Florida State University campus, received over six inches of rain during the event.

The satellite-borne Special Sensor Microwave Imager (SSM/I) is a seven channel, four frequency, linearly polarized, passive microwave radiometric system. It measures the upwelling scene brightness temperatures of the earth's surface, to include the brightness temperatures of hydrometeor concentrations (clouds). The Defense Meteorological Satellite Program (DMSP) designs, builds and launches the polar orbiting satellites which carry the SSM/I system. Three satellites are currently operational (F11, F13, F14), and for this study we used data from all three sources.

Meteorological data are sparse on a global scale, especially over the oceans, so the SSM/I coverage has provided a new and valuable global data source. Since the late 1980's, various computer algorithms have been developed to calculate global rain rates from the SSM/I brightness temperature data. These estimated rain rate data can be used both in global weather prediction models and by meteorologists forecasting downstream of data-sparse regions. At Florida State, Dr. T.N. Krishnamurti's laboratory utilizes SSM/I rain rates in the physical initialization of the FSU Global Spectral Model. Using a combination of rain rate data from ground-based observations, outgoing longwave radiation (OLR), and SSM/I derived rain rates, this model ultimately initializes the physical structure of tropical cyclones. Consequently, the accuracy of SSM/I rain rate estimations is an exceedingly important ingredient in the model prediction of hurricane intensity and movement.

This paper will serve as a short and specific validation and comparison study of three separate Special Sensor Microwave Imager (SSM/I) rain rate algorithms, using the Tallahassee official observation of 4.43 inches as the baseline. The algorithms used in this case study to compute SSM/I rain rate estimates from raw SSM/I data are commonly referred to as CALVAL, NESDIS, and GPROF. The purpose of this paper is to highlight which of these algorithms generates the most accurate rain rate estimate. It is important to note that rain rates depicted in the attached figures are averaged over 24 hours in units of millimeters.

Algorithm Descriptions

Both the CALVAL and NESDIS SSM/I rain rate algorithms can be described as statistical rain maps. These two algorithms derive a statistical regression between a measured brightness temperature data set involving one or more frequencies, and a rainfall data set obtained from rain gauge and/or radar ground measurements (Smith et al., 1998).

The GPROF rain rate algorithm, on the other hand, is a physical profile algorithm that retrieves the vertical structure of one or more hydrometeor categories through multispectral inversion rather than just surface rain rates (Smith et al., 1998).


The CALVAL algorithm is the United States Navy's second generation precipitation retrieval algorithm. Developed by the Calibration/Validation group in the early 1990's, it replaced the Navy's D-Matrix algorithm for operational use. The CALVAL algorithm is based upon the statistical regression of SSM/I brightness temperatures against collocated surface radar measurements of rain rate (Berg et al., 1998). This algorithm is described by Olson et al. (1991) in volume II of the CALVAL Final Report (CalVal 1991).

Training data sets from radar-derived rain rates created at Kwajalein atoll and Darwin, Australia, were used to develop this algorithm. The radar-derived rain rates were used in a regression expression relating the log of the rain rate to five of the seven SSM/I brightness temperatures over the ocean, but only to the 85 GHz vertical and horizontal brightness temperatures for land areas (Smith et al., 1998).

The results of this case study for Tallahassee, Florida, are later shown to be in line with recent follow-up studies such as the Precipitation Intercomparison Project (PIP-2). These studies confirmed a scarcity of high intensity rain events in the training data sets. Consequently, a bias towards underestimation exists in the algorithm where rain rates do not exceed 6 mm/hr. In addition, the CALVAL algorithm has been documented to detect overly broad regions of precipitation (Berg et al., 1998).


The NESDIS SSM/I rain rate algorithm is the Navy's third generation algorithm developed by the National Environmental Satellite, Data and Information System (Ferraro and Marks 1995). The algorithm used in this case study is the most recent version of NESDIS, which includes an emission component (older versions only used the 85 GHz scattering).

The NESDIS algorithm bases its rain rates on non-linear regression between the 85 GHz scattering index used in a screening procedure and operational radar measurements (Ferraro et al., 1994; Ferraro and Marks 1995). The radar data are obtained from three worldwide sites to include the 22 platform Japanese Meteorological Agency AMeDAS system, the 14 platform U.K. Meteorological Office FRONTIERS system, and the 13 platform U.S. National Weather Service RADAP-II system (Smith et al., 1998).

Similar in nature to the CALVAL algorithm, the NESDIS algorithm separates land and ocean areas. This distinction is in response to the greater microwave emissivity of land surfaces compared to ocean surfaces. Power laws are used in the ocean - land conversion relationship, with exponents for the scattering indexes of approximately 2.0. Since the conversion of the 85 GHz scattering index to a rain rate uses a power fit, rapid increases can occur for extreme rain events. Consequently, the maximum rain rate is bounded at 35 mm/hr.

The NESDIS algorithm has recently been updated to include many improvements from its original version. The algorithm currently incorporates enhanced sensitivity to oceanic scattering plus improved screening for anomalous scattering features such as frozen ground and non-vegetated land. While the older version used antenna temperatures, the latest version uses coefficients derived for brightness temperatures (Ferraro and Marks 1995).


The Goddard Profiling (GPROF) SSM/I rain rate algorithm uses statistical inversion techniques based upon theoretically calculated relations between rainfall rates and brightness temperatures. Through the explicit accounting of diverse hydrometeor profiles, potential errors introduced into the GPROF theoretical calculations by the unknown vertical hydrometeor distribution are overcome. This is accomplished by allowing various vertical distributions in the theoretical brightness temperature calculations and by requiring consistency between the observed and calculated brightness temperatures (Kummerow and Giglio 1994). Smith et al. (1998) states that the cloud model generated profiles in the GPROF algorithm are assigned a priori probabilities and then related to brightness temperatures at all SSM/I frequencies and polarizations through a forward RTE model.

The GPROF SSM/I rain rate algorithm explicitly accounts for the vertical structure of precipitation within a cloud. This vertical structure is a key factor in determining the upwelling radiances, and the algorithm simulates the upwelling radiances by means of a radiative transfer scheme (Kummerow and Giglio 1994). GPROF 4.0 is the version used for this case study. Separate versions of the GPROF algorithm exist for use in the Tropical Rainfall Measuring Mission (TRMM).


Figure 1 shows rain rate estimates calculated by the CALVAL algorithm from southeast Louisiana to south Florida for July 13th, 1998 (Julian day 194). The plotted contours indicate a maximum rain rate of 20 mm/day in the vicinity of Tallahassee, Florida (30.5 degrees north latitude, 84.5 degrees west longitude).

Figure 2 shows rain rate estimates calculated by the NESDIS algorithm for the same region and day. The maximum rain rate estimate generated in the vicinity of Tallahassee is 80 mm/day.

Figure 3 shows rain rate estimates calculated by the GPROF 4.0 SSM/I rain rate algorithm. The GPROF algorithm estimates a maximum rain rate of 40 mm/day in Tallahassee for July 13th, 1998.


It is clearly evident that for purposes of this case study, in which the objective is to validate the accuracy of SSM/I estimated rain rates using a specific ground based observation, the NESDIS algorithm provides the most accurate rain rate estimate. Using the conversion of 1 inch equals 25 millimeters, the official Tallahassee observation of 4.43 inches is equivalent to 110.75 millimeters. The NESDIS estimate of 80 mm/day is the only estimate of the three algorithms which comes close to the official 24 hour observation of 110.75 millimeters.

Both the CALVAL and GPROF algorithms grossly underestimate the magnitude of precipitation which reached the ground in Tallahassee. In addition, the CALVAL algorithm shows an inaccurate departure in the geographical extent of the rain event in question. While the NESDIS and GPROF algorithms accurately reflect the concentrated downpour of rain which occurred in Tallahassee, the CALVAL algorithm spreads the rain event over a broader region, geographically as far as central Alabama and central Georgia.

The performances of the three algorithms in question are specific to the Tallahassee based thunderstorm event from July 13th, 1998 only. Beyond the limited scope of this study, these SSM/I algorithms can exhibit other strengths and weaknesses in depicting rain rate estimates in other areas of the world, most specifically in coastal and oceanic locations. Therefore this paper does not attempt to provide a general, all-inclusive statement concerning algorithm accuracy for any other rain and/or thunderstorm event(s).


I would like to thank Professor T. N. Krishnamurti for his leadership, ideas, and suggestions in support of this paper. Major Ed Bensman deserves recognition for having an answer to every question I have asked in the past months. I would also like to thank Mr. Eric Williford and Mr. Curtis Knox for keeping the computer resources operational. In addition, I would like to thank Mr. Ricardo Correa-Torres for his assistance with manipulating the GRADS software. And finally I would like to recognize Rosemarie Raymond for her skill in preparing the graphics.


Berg, W., W. Olson, R. Ferraro, S.J. Goodman, and F.J. LaFontaine, 1998: An Assessment of the First- and Second-Generation Navy Operational Precipitation Retrieval Algorithms. J. Atmos. Sci., 55, 1558-1575.

Ferraro, R. R., and G. F. Marks, 1995: The Development of SSM/I Rain-Rate Retrieval Algorithms Using Ground-Based Radar Measurements. J. Atmos. Oceanic Technol., 12, 755-770.

Kummerow, C., L. Giglio, 1994: A Passive Microwave Technique for Estimating Rainfall and Vertical Structure Information from Space. Part I: Algorithm Description. J. Appl. Meteor., 33, 3-18.

Smith, E. A., J. E. Lamm, R. Adler, J. Alishouse, K. Aonashi, E. Barrett, P. Bauer, W. Berg, A. Chang, R. Ferraro, J. Ferriday, S. Goodman, N. Grody, C. Kidd, D. Kniveton, C. Kummerow, G. Liu, F. Marzano, A. Mugnai, W. Olson, G. Petty, A. Shibata, R. Spencer, F. Wentz, T. Wilheit, and E. Zipser, 1998: Results of WetNet PIP-2 Project. J. Atmos. Sci., 55, 1483-1536.